Cohort retention: curve, churn and cohort LTV
For cohorts of customers (the month of the first purchase) it builds the retention curve: what share came back in the 1st, 2nd, 3rd and 4th period after the first purchase. It compares cohorts by LTV over a common horizon and shows which cohort brought more valuable customers.
At a glance
- What question does it answer?
- How fast cohorts lose customers and which cohort brought more valuable customers.
- What data is needed?
- By cohort (the month of the first purchase): the size and the number active in periods 1–4; empty if the period has not come.
- How to read the result?
- On the example of three cohorts retention in period 1 is 32.7% (980 of 3,000 customers) and in period 4 it is 13% (from the one cohort with such data). The common horizon of the cohorts is 2 periods: over it LTV at a margin per order of 1,500 and 1.3 orders per active customer is 2,553 on average, 2,637.5 for the March cohort (the best) and 2,450.6 for the February cohort (the worst), a gap of 7.6%. If the January cohort over 4 periods (3,099) were compared with the others over 2, it would look the best: this is the error made when horizons differ. Example on the default values: Retention in period 1 (weighted): 32.7%; Retention in the last observed period (4): 13%; LTV over a horizon of 2 periods (cohort average): RUB 2,553; LTV of the best cohort over the same horizon: RUB 2,638 · March cohort. The figures are illustrative: enter your own to see your result.
Why and how to reason
Average retention across the whole base mixes old and new customers and hides the main point: did new customers start to behave better or worse. A cohort is customers who made a first purchase in the same month; we follow what share returns after 1, 2, 3 periods. Retention in period i = Σ actives in period i ÷ Σ customers of the cohorts where that period has already occurred (immature cohorts do not distort). In the example the January cohort of 1,000 retains 32%, 21%, 16%, 13%; weighted retention in period 1 is 980 ÷ 3,000 = 32.7%. LTV = order margin × (1 + orders per active × Σ retentions over the horizon). On the common horizon of 2 periods for the March cohort: 1,500 × (1 + 1.3 × (0.35 + 0.233)) = 2,637, for February 2,451: the best is 7.6% above the worst. Comparison is possible only on the same horizon.
When to use it
Reach for it to judge acquisition quality: channel, promotion, collection, new policy. It fits when you have purchase history for at least 3–4 periods. It does not fit rare purchases (once a season) over short periods and small cohorts (dozens of customers): shares jump. Assumptions: order margin and frequency are constant across periods; enter CAC if you want to see cohort payback.
How to read the result
Look at the shape of the curve: a sharp drop after the first period and then a plateau means you lose casual buyers but the core stays. A curve that falls to zero means there are no repeat purchases and the business needs constant inflow. Compare cohorts on the same horizon: if a new cohort is weaker than old ones, something changed (channel, discounts, assortment). LTV to CAC above 1 means the cohort paid back; set the benchmark yourself from the time in which you are ready to recoup money. The decision: reallocate budget to the channel with the better cohort.
Common mistakes
Comparing cohorts on different horizons: the old one looks better simply because it was observed longer. Counting as gone those who will simply buy later. Mixing customers of different channels in a cohort: the average hides the good and the bad channel. Also do not take one cohort’s jump for a trend: check the cohort size and the season it arrived in.
What data is needed?
- By cohort (the month of the first purchase): the size and the number active in periods 1–4; empty if the period has not come.
- Margin per order excl. VAT and orders per active customer; optionally CAC.
Inputs
| Field | Unit | Value in the example | Meaning |
|---|---|---|---|
| Margin per order excl. VAT | ₽ | 1,500 | Gross margin of an average order after discounts. Amount excl. VAT: the general rate is 22% from 01.01.2026. The rate follows the shipment date; confirm your regime and rate with an accountant. |
| Orders per active customer per period | × | 1.3 | How many orders in a period a customer who bought in it makes. Take it from your own data. |
| Customer acquisition cost (CAC) | ₽ | — | Optional. Advertising and other acquisition costs ÷ new customers. Empty: no comparison with CAC. |
| Retention tail beyond 4 periods | No: only the observed periods | Tail: retention after the last observed period falls by the given percent per period. This is an assumption, not data: check it on old cohorts. | |
| Periods of the tail | periods | 8 | How many periods to extend the curve. |
| Fall of retention in the tail per period | % | 20 | An illustrative example: take the ratio of retention of neighboring periods from your data. |
| Margin layers by cohort | Do not show | Cohort margin in a period: the first order in period 0 and active customers × orders per active customer × margin in periods 1–4. This is margin, not revenue. | |
| “Churned” threshold: the period without a purchase | periods | — | Optional. The period in which a customer who has not bought is counted as churned (from the “Churn threshold” calculation on your purchase intervals). Empty: not counted. |
How it is calculated
| Step | Formula |
|---|---|
| Retention in period i | Σ active in period i ÷ Σ customers of the cohorts where period i has come |
| LTV | margin per order × (1 + orders per active customer × Σ retentions over the horizon) |
| Common horizon | the number of first periods observed in all cohorts |
Worked numbers
Default values from the app; in the interactive version you enter your own.
- Retention in period 1 (weighted)
- 32.7 %
- Retention in the last observed period (4)
- 13 %
- LTV over a horizon of 2 periods (cohort average)
- 2,553 ₽Margin per order × (1 + orders per active customer × sum of retentions).
- LTV of the best cohort over the same horizon
- 2,638 ₽ · March cohort
- LTV of the worst cohort over the same horizon
- 2,451 ₽ · February cohort
- The best is above the worst by
- 7.6 %
- LTV minus CAC
- enter your own CAC
- LTV to CAC
- enter your own CAC
- 18.8 … 35 %
Data as a table
| Period 1 | Period 2 | |
|---|---|---|
| January cohort | 32 | 21 |
| February cohort | 30 | 18.8 |
| March cohort | 35 | 23.3 |
Data as a table
| Measure | Retention |
|---|---|
| Period 1 | 32.7 |
| Period 2 | 21.3 |
| Period 3 | 15 |
| Period 4 | 13 |
Retention and LTV by cohort
| Cohort | Customers | Period 1 | Period 2 | Period 3 | Period 4 | LTV (2 per.) |
|---|---|---|---|---|---|---|
| January cohort | 1,000 | 32 % | 21 % | 16 % | 13 % | 2,534 ₽ |
| February cohort | 800 | 30 % | 18.8 % | 13.8 % | — | 2,451 ₽ |
| March cohort | 1,200 | 35 % | 23.3 % | — | — | 2,638 ₽ |
Cohorts are compared over the common horizon of 2 period(s): a younger cohort does not lose just because it has less data.
In periods after the common horizon the average covers only the cohorts where they have already come; that is a different set of customers.
There is no single retention standard in Russia; compare the curve with your own past cohorts. Margin and orders per customer are constant here: a simplification.
Nuances and limits
- Comparing cohorts over a different number of periods.
- Counting as “active in the period” those who bought at least once after the first purchase: that is a cumulative share, not retention.
- There is no single retention standard in Russia: compare with your own cohorts.